harfbuzz/docs/usermanual-shaping-concepts.xml

<chapter id="shaping-concepts">
  <title>Shaping concepts</title>
  <section id="text-shaping-concepts">
    <title>Text shaping</title>
    <para>
      Text shaping is the process of transforming a sequence of Unicode
      codepoints that represent individual characters (letters,
      diacritics, tone marks, numbers, symbols, etc.) into the
      orthographically and linguistically correct two-dimensional layout
      of glyph shapes taken from a specified font.
    </para>
    <para>
      For some writing systems (or <emphasis>scripts</emphasis>) and
      languages, the process is simple, requiring the shaper to do
      little more than advance the horizontal position forward by the
      correct amount for each successive glyph.
    </para>
    <para>
      But, for <emphasis>complex scripts</emphasis>, any combination of
      several shaping operations may be required, and the rules for how
      and when they are applied vary from script to script. HarfBuzz and
      other shaping engines implement these rules.
    </para>
    <para>
      The exact rules and necessary operations for a particular script
      constitute a shaping <emphasis>model</emphasis>. OpenType
      specifies a set of shaping models that covers all of
      Unicode. Other shaping models are available, however, including
      Graphite and Apple Advanced Typography (AAT). 
    </para>
  </section>
  
  <section id="complex-scripts">
    <title>Complex scripts</title>
    <para>
      In text-shaping terminology, scripts are generally classified as
      either <emphasis>complex</emphasis> or <emphasis>non-complex</emphasis>.
    </para>
    <para>
      Complex scripts are those for which transforming the input
      sequence into the final layout requires some combination of
      operations&mdash;such as context-dependent substitutions,
      context-dependent mark positioning, glyph-to-glyph joining,
      glyph reordering, or glyph stacking.
    </para>
    <para>
      In some complex scripts, the shaping rules require that a text
      run be divided into syllables before the operations can be
      applied. Other complex scripts may apply shaping operations over
      entire words or over the entire text run, with no subdivision
      required.
    </para>
    <para>
      Non-complex scripts, by definition, do not require these
      operations. However, correctly shaping a text run in a
      non-complex script may still involve Unicode normalization,
      ligature substitutions, mark positioning, kerning, and applying
      other font features. The key difference is that a text run in a
      non-complex script can be processed sequentially and in the same
      order as the input sequence of Unicode codepoints, without
      requiring an analysis stage.
    </para>
  </section>
  
  <section id="shaping-operations">
    <title>Shaping operations</title>
    <para>
      Shaping a complex-script text run involves transforming the
      input sequence of Unicode codepoints with some combination of
      operations that is specified in the shaping model for the
      script.
    </para>
    <para>
      The specific conditions that trigger a given operation for a
      text run varies from script to script, as do the order that the
      operations are performed in and which codepoints are
      affected. However, the same general set of shaping operations is
      common to all of the complex-script shaping models. 
    </para>
    
    <itemizedlist>
      <listitem>
	<para>
	  A <emphasis>reordering</emphasis> operation moves a glyph
	  from its original ("logical") position in the sequence to
	  some other ("visual") position.
	</para>
	<para>
	  The shaping model for a given complex script might involve
	  more than one reordering step.
	</para>
      </listitem>
      
      <listitem>
	<para>
	  A <emphasis>joining</emphasis> operation replaces a glyph
	  with an alternate form that is designed to connect with one
	  or more of the adjacent glyphs in the sequence.
	</para>
      </listitem>
      
      <listitem>
	<para>
	  A contextual <emphasis>substitution</emphasis> operation
	  replaces either a single glyph or a subsequence of several
	  glyphs with an alternate glyph. This substitution is
	  performed when the original glyph or subsequence of glyphs
	  occurs in a specified position with respect to the
	  surrounding sequence. For example, one substitution might be
	  performed only when the target glyph is the first glyph in
	  the sequence, while another substitution is performed only
	  when a different target glyph occurs immediately after a
	  particular string pattern.
	</para>
	<para>
	  The shaping model for a given complex script might involve
	  multiple contextual-substitution operations, each applying
	  to different target glyphs and patterns, and which are
	  performed in separate steps.
	</para>
      </listitem>
      
      <listitem>
	<para>
	  A contextual <emphasis>positioning</emphasis> operation
	  moves the horizontal and/or vertical position of a
	  glyph. This positioning move is performed when the glyph
	  occurs in a specified position with respect to the
	  surrounding sequence.
	</para>
	<para>
	  Many contextual positioning operations are used to place
	  <emphasis>mark</emphasis> glyphs (such as diacritics, vowel
	  signs, and tone markers) with respect to
	  <emphasis>base</emphasis> glyphs. However, some complex
	  scripts may use contextual positioning operations to
	  correctly place base glyphs as well, such as
	  when the script uses <emphasis>stacking</emphasis> characters.
	</para>
      </listitem>
      
    </itemizedlist>
  </section>
  
  <section id="unicode-character-categories">
    <title>Unicode character categories</title>
    <para>
      Shaping models are typically specified with respect to how
      scripts are defined in the Unicode standard.
    </para>
    <para>
      Every codepoint in the Unicode Character Database (UCD) is
      assigned a <emphasis>Unicode General Category</emphasis> (UGC),
      which provides the most fundamental information about the
      codepoint: whether the codepoint represents a
      <emphasis>Letter</emphasis>, a <emphasis>Mark</emphasis>, a
      <emphasis>Number</emphasis>, <emphasis>Punctuation</emphasis>, a
      <emphasis>Symbol</emphasis>, a <emphasis>Separator</emphasis>,
      or something else (<emphasis>Other</emphasis>).
    </para>
    <para>
      These UGC properties are "Major" categories. Each codepoint is
      further assigned to a "minor" category within its Major
      category, such as "Letter, uppercase" (<literal>Lu</literal>) or
      "Letter, modifier" (<literal>Lm</literal>).
    </para>
    <para>
      Shaping models are concerned primarily with Letter and Mark
      codepoints. The minor categories of Mark codepoints are
      particularly important for shaping. Marks can be nonspacing
      (<literal>Mn</literal>), spacing combining
      (<literal>Mc</literal>), or enclosing (<literal>Me</literal>).
    </para>
    <para>
      In addition to the UGC property, codepoints in the Indic and
      Southeast Asian scripts are also assigned
      <emphasis>Unicode Indic Syllabic Category</emphasis> (UISC) and
      <emphasis>Unicode Indic Positional Category</emphasis> (UIPC)
      property that provides more detailed information needed for
      shaping.
    </para>
    <para>
      The UISC property sub-categorizes Letters and Marks according to
      common script-shaping behaviors. For example, UISC distinguishes
      between consonant letters, vowel letters, and vowel marks. The
      UIPC property sub-categorizes Mark codepoints by the visual
      position that they occupy (above, below, right, left, or in
      multiple positions).
    </para>
    <para>
      Some complex scripts require that the text run be split into
      syllables, and what constitutes a valid syllable in these
      scripts is specified in regular expressions of the Letter and
      Mark codepoints that take the UISC and UIPC properties into account.
    </para>

  </section>
  
  <section id="text-runs">
    <title>Text runs</title>
    <para>
      Real-world text usually contains codepoints from a mixture of
      different Unicode scripts (including punctuation, numbers, symbols,
      white-space characters, and other codepoints that do not belong
      to any script). Real-world text may also be marked up with
      formatting that changes font properties (including the font,
      font style, and font size).
    </para>
    <para>
      For shaping purposes, all real-world text streams must be first
      segmented into runs that have a uniform set of properties. 
    </para>
    <para>
      In particular, shaping models always assume that every codepoint
      in a text run has the same <emphasis>direction</emphasis>,
      <emphasis>script</emphasis> tag, and
      <emphasis>language</emphasis> tag.
    </para>
  </section>
  
  <section id="opentype-shaping-models">
    <title>OpenType shaping models</title>
    <para>
      OpenType provides shaping models for the following scripts:
    </para>

    <itemizedlist>
      <listitem>
	<para>
	  The <emphasis>default</emphasis> shaping model handles all
	  non-complex scripts, and may also be used as a fallback for
	  handling unrecognized scripts.
	</para>
      </listitem>

      <listitem>
	<para>
	  The <emphasis>Indic</emphasis> shaping model handles the Indic
	  scripts Bengali, Devanagari, Gujarati, Gurmukhi, Kannada,
	  Malayalam, Oriya, Tamil, Telugu, and Sinhala.
	</para>
	<para>
	  The Indic shaping model was revised significantly in
	  2005. To denote the change, a new set of <emphasis>script
	  tags</emphasis> was assigned for Bengali, Devanagari,
	  Gujarati, Gurmukhi, Kannada, Malayalam, Oriya, Tamil, and
	  Telugu. For the sake of clarity, the term "Indic2" is
	  sometimes used to refer to the current, revised shaping
	  model.
	</para>
      </listitem>

      <listitem>
	<para>
	  The <emphasis>Arabic</emphasis> shaping model supports
	  Arabic, Mongolian, N'Ko, Syriac, and several other connected
	  or cursive scripts.
	</para>
      </listitem>

      <listitem>
	<para>
	  The <emphasis>Thai/Lao</emphasis> shaping model supports
	  the Thai and Lao scripts.
	</para>
      </listitem>

      <listitem>
	<para>
	  The <emphasis>Khmer</emphasis> shaping model supports the
	  Khmer script.
	</para>
      </listitem>

      <listitem>
	<para>
	  The <emphasis>Myanmar</emphasis> shaping model supports the
	  Myanmar (or Burmese) script.
	</para>
      </listitem>

      <listitem>
	<para>
	  The <emphasis>Tibetan</emphasis> shaping model supports the
	  Tibetan script.
	</para>
      </listitem>

      <listitem>
	<para>
	  The <emphasis>Hangul</emphasis> shaping model supports the
	  Hangul script.
	</para>
      </listitem>

      <listitem>
	<para>
	  The <emphasis>Hebrew</emphasis> shaping model supports the
	  Hebrew script.
	</para>
      </listitem>

      <listitem>
	<para>
	  The <emphasis>Universal Shaping Engine</emphasis> (USE)
	  shaping model supports complex scripts not covered by one of
	  the above, script-specific shaping models, including
	  Javanese, Balinese, Buginese, Batak, Chakma, Lepcha, Modi,
	  Phags-pa, Tagalog, Siddham, Sundanese, Tai Le, Tai Tham, Tai
	  Viet, and many others. 
	</para>
      </listitem>

      <listitem>
	<para>
	  Text runs that do not fall under one of the above shaping
	  models may still require processing by a shaping engine. Of
	  particular note is <emphasis>Emoji</emphasis> shaping, which
	  may involve variation-selector sequences and glyph
	  substitution. Emoji shaping is handled by the default
	  shaping model.
	</para>
      </listitem>

    </itemizedlist>
    
  </section>
  
  <section id="graphite-shaping">
    <title>Graphite shaping</title>
    <para>
      In contrast to OpenType shaping, Graphite shaping does not
      specify a predefined set of shaping models or a set of supported
      scripts.
    </para>
    <para>
      Instead, each Graphite font contains a complete set of rules that
      implement the required shaping model for the intended
      script. These rules include finite-state machines to match
      sequences of codepoints to the shaping operations to perform.
    </para>
    <para>
      Graphite shaping can perform the same shaping operations used in
      OpenType shaping, as well as other functions that have not been
      defined for OpenType shaping.
    </para>
  </section>
  
  <section id="aat-shaping">
    <title>AAT shaping</title>
    <para>
      In contrast to OpenType shaping, AAT shaping does not specify a 
      predefined set of shaping models or a set of supported scripts.
    </para>
    <para>
      Instead, each AAT font includes a complete set of rules that
      implement the desired shaping model for the intended
      script. These rules include finite-state machines to match glyph
      sequences and the shaping operations to perform.
    </para>
    <para>
      Notably, AAT shaping rules are expressed for glyphs in the font,
      not for Unicode codepoints. AAT shaping can perform the same
      shaping operations used in OpenType shaping, as well as other
      functions that have not been defined for OpenType shaping.
    </para>
  </section>
</chapter>